PosWSD: Low-Resource Word Sense Disambiguation Model using Part Of Speech Information
Yazhen Chen, Jian Zhang, Qipeng He · 2022
Word Sense Disambiguation(WSD) is a long-standing problem in Natural Language Processing(NLP), which aims to find the exact meaning of the target word in a given context. Current WSD methods mainly rely on pre-trained models. They use large-scale annotated data to take advantage of additional knowledge for disambiguation. Due to the limitations of annotated data and open source knowledge base, current WSD methods mainly disambiguate the data of WordNet annotation system. For the situation where there is no open source knowledge base, little annotated data and few computational resources, we propose PosWSD: a WSD model using part of speech(POS) information. The model learns the context and candidate word meanings based on the idea of treating the Word Sense Disambiguation task as a binary task, and then introduces POS of the context which improves its performance. It can be used for the English WSD task of non-WordNet annotation system. The experimental results show that incorporating POS information is necessary for the Word Sense Disambiguation task, which is beneficial for the model to understand the semantics.